High. Lithography Systems. * prof. Dr. A[ec N. Broers, ZBM T. J. Watson Re- printing is that the close proximity between

Transcription

1 High Resol ution Lithography Systems A Review of the Current Status By Alec N. Broers, Yorktown Heights, N.Y.*) This paper discusses the advantages and disadvantages of advanced lithography techniques under investigation for the fabrication of semiconductor integrated circuits. These techniques are replacing conventional ultraviolet (UV) contact/proximity printing. They employ standard UV radiation (A = 3500 A-4000 A), deep UV radiation (1 = 2000 A-2600 A), soft x-rays (A = 4 A--40 A), and electrons. It is assumed that the reader is familiar with the basic principles of the new methods. veloped which are capable of reaching dimensions close to 1 pm and 2) recognized problems with the new technologies, for example, cost for scanning electron beams systems, and throughput and mask stability for x-ray lithography are taking longer to resolve than optimistic proponents of these methods had originally predicted. The situation is further complicated by the proliferation of available approaches. At present there are eight different methods under investigation for dimensions below 2 pm. They are contact/proximity printing with deep UV radiation, x-ray lithography, 1 : 1 scanning optical projection, step and repeat reduction optical projection, scanning electron beam, 1 : 1 photocathode electron beam projection, step and repeat reduction electron beam projection, and proximity printing with electrons. This paper discusses the advantages and disadvantages of these methods. It is assumed that the reader is familiar with the principles of the different techniques. They have recently been reviewed elsewhere [I]. A few representative references are included to give the reader a path into the extensive literature on the subject. Contact/Proximity Methods Conventional UV proximity printing methods can be extended to resolutions below Introduction 2 pm through the use of deep UV (A = 2000 A-2500 A) radiation [2](see for example Lithography in this context is the art of figure 1). Further improvement to below defining the intricate patterns needed for the fabrication of microcircuits. For dimensions below about 2.5 pm it used to be thought that light optical processes would to be abandonned and electron beam or x-ray methods adopted. This was when there was little interest in reduced dimensions in the microcircuit industry. Now this interest is widespread, but the situation has changed in two important ways: 1) optical systems have been de- 1 pm is available through the simultaneous use of conformal printing methods 131 in which the gap between mask and wafer is minimized. Alignment techniques are available which can in principle reach accuracies of 0.25 pm. These combined capabilities give contact/proximity printing the potential to reach resolutions beyond those of present 1 x optical projection systems. Throughput can be comparable and system cost is lower. The reason not to use contact/proximity * prof. Dr. A[ec N. Broers, ZBM T. J. Watson Re- printing is that the close proximity between search Center, P.O. Box 218, Yorktown Heights, N.Y , USA. 704 mask and wafer inevitably results in mask damage which reduces yield and increases

2 the number of masks required for a given wafer volume. This disadvantage has clearly justified the change to projection printing at 3-4 pm linewidths. The situation should be reexamined, however, in light of the potential density improvement achievable at linewidths of 1-2 pm. Figure 4. Proximity printed image using deep UV (2000 A) radiation in PMMA resist. 5,urn gap was maintained between mask and wafer (Lin [21). Linewidth 1.5,urn. Figure 2. High aspect ratio pattern in PMMA resist obtained by x-ray lithography (Spiller et. al., J. de Physique, 1978, to be published). 1,urn linewidths. X-Ray Lithography X-ray lithography [4, 51 (proximity printing with soft x-rays) offers one outstanding advantage over other methods. Image fidelity is preserved in very thick resist layers and is not degraded by diffraction effects as it is with light optical exposure, or scattering as it is with electron exposure. This allows high aspect ratio resist patterns to be produced (see figure l), and means that there are no proximity effects as there are with electron exposure. On the other hand, the mask is fragile and may not be dimensionally stable, and the large divergence of conventional x- ray sources gives rise to image distortion if the mask or wafer are not flat. A step and repeat x-ray approach in which the mask size is limited to the maximum stable area (this area can be set by mask and/or wafer instabilities) overcomes the problems of mask stability and source divergence, but exposure time with present resists and sources becomes excessive. A synchrotron storage ring [61 with a circulating current of at least tens of milliamps is a bright enough source to reduce exposure times to an acceptable level, but its great cost and overcapacity makes it difficult to utilize efficiently. The most satisfactory solution to this set of problems would appear to be the development of a compact and relatively inexpensive high output soft x-ray source, or a significant improvement in resist sensitivity. Recent experiments using x-rays from laser produced plasmas [71 offer some hope for the former. For the latter, it must be remembered that sensitivity has to be defined in terms of the resist (wall) profile and the acceptable thickness loss in the resist areas that remain after development (8). These requirements generally become more stringent for smaller dimensions, and resist processes that have been adequate for larger dimensions may no longer be useful. Resists that have proved to be satisfactory for 1 pm device fabrication do not have adequate sensitivity for the step and repeat approach unless a storage ring is used as the x-ray source. 705

3 The alternative to the step and repeat approach. is the original concept of full-wafer printing. Here a resilient substrate with adequate dimensional stability must be found. The formidable technological challenges must also be met of controlling temperature stability, and sample and mask chucking forces to the point that 2 ppm overall dimensional repeatability (0.2 pm over 10 cm) is achieved. Many laboratories are concentrating their efforts on these problems which enhances the likelihood of their being resolved. Experimental results already indicate that alignment of mask and wafer should be possible to better than 0.1 pm 191. I x Scanning Optical Projection 1 : 1 optical scanning projection systems I101 use the same masks and resists as contact printing, but they improve yield greatly by keeping mask and wafer separated. They also utilize master masks and thus dispense with the need for working masks. Throughput is high and approaches one hundred 8 cm diameter wafers per hour in some instances [lo]. Resolution at present is adequate for 3 pm linewidths, and new systems promise improvement to 2 pm or below. Overlay accuracy is predicted to improve from + 1 pm -1.5 pm to pm. In the new systems tighter tolerances are met in the fabrication of optical components, temperature corprol is improved, and deep UV (A = 2000 A-2600 A) radiation may be used, rather than conventional wavelengths (A = 3000 A-4000 A). I x scanning systems are full-wafer exposure systems and will encounter severe overall dimensional stability problems as dimensions approach 1 pm. The numerical aperture of the mirror imaging system will also have to be increased from the present value of 0.16 to about 0.3, and even with partially coherent illumination, depth of field will become very small. This may be intolerable because it will be very difficult to adjust focus and level the sample for different portions of the sample surface. Figure 3. 1 Am linewidths in resist produced by Sx projection camera (Wilczynski. J. S.. Tibbetts, R., Lin. B. J.. and Santy, W., IBM Research, Yorktown Heights, N. Y.). Step and Repeat Reduction Projection Step and repeat reduction projection cameras offer the same yield advantage as 1 :1 scanning projection plus a demonstrated improvement in resolution. High quality lenses have been made which achieve diffraction limited resolution at a numerical aperture of 0.34 N.A. (60 o/o contrast for 1.25 pm lines) over a field of I cm x 1 cm Advances in the design of illumination systems allow similar fields to be exposed in less than 0.5 sec (see for example 1131). When such an illumination system is combined with a high speed interferometrically controlled table, throughputs of 50 75mm diameter wafers per hour can be obtained Throughput had previously been considered inadequate for many LSI applications. As already mentioned, a major difficulty with any optical system at resolutions below 2 pm is the very small depth of field. In the step and repeat case, however, focusing and wafer leveling can be carried out at every chip site and the problem more readily contained than with the full wafer scanning systems. Even with this adjustment resist pro- 106

4 cesses have to be very closely controlled and multilevel resists in which the imaging layer is kept thin will be particularly advantageous. Alignment can be carried out once per wafer, as with 1 : 1 scanning systems, or preferably at every chip site. In the first instance laser interferometry is relied upon to maintain alignment and overlay accuracy of rt 0.5 pm should be achievable. In the second case improvement to? 0.25 pm should be possible. Although the resolution (see figure 3 and reference 15), throughput and overlay accuracy are very encouraging for step and repeat cameras, full complexity LSI devices with dimensions below 2 pm have yet to be fabricated. The most serious unknown is the ability to maintain pattern fidelity on rough surfaces. Scanning Electron Beam This is the only method used so far to successfully make fullcomplexity silicon devices with a capability that exceeds conventional contact printing in terms of linewidth and overlay tolerance. See, for example, devices described in references [16] and [171. Full lithography capability has been established for 0.6 pm dimensions as is shown in figure 4. Masks made by a scanning electron beam Figure pm gate length F.E.T. test chip fabricated by VS-1 scanning electron beam system (Chang, T. H. P., et al., Proc. 7th Internat. Sym. Electron Ion Beam Sci. Technol. (Electrochem. Soc. Princeton, N. J.), p. 392, (1976).) are also needed for replication methods such as x-ray lithography and deep UV conformal printing. For conventional masks, scanning electron beam systems offer better resolution, lower defect levels, and quicker turn-round than conventional mechanical pattern generation techniques [ An electron beam system can generate several master masks for 8 cm diameter samples in less than 1 hour, whereas it can take several hours to generate the reticule alone with standard methods. This reticule then has to be stepped and repeated in a reduction projection camera to produce a master mask. The quick turn-round of scanning beam fabrication has also been used to advantage for the direct writing of customized interconnection layers on LSI circuits [221. Although scanning electron beam systems have proved that they are capable of directly fabricating advanced circuits, they are not yet cost competitive with optical approaches for the manufacturing of most LSI devices. Nonetheless, their performance in this respect is improving rapidly. The improvement is due to the introduction of advanced electron optical design methods [231 and concepts (for example, the variable shaped beam concept [24, 251), higher speed electronics, and due to large reductions in the cost of the digital control electronics. Efforts are also being made to evolve optimum system architectures. In particular, the possibility of minimizing overhead times spent on mechanical sample movement and registration has become apparent through a combination of large area two dimensional electronic scanning, vector pattern addressing, and a fast variable speed mechanical table. The table would be moved under the control of a laser interferometer without interrupting the pattern writing process [261. Resolution in electron beam fabricated structures is not ultimately set by the resolution of the electron optical system, but by electron scattering in the image-forming layer (resist), and by electron backscattering from the substrate. These scattering pheno- I01

5 mena also give rise to proximity effects which make it necessary to adjust the exposure for different areas of the pattern. Experiments with automatic proximity effect correction indicate that correction is essential for dimensions below 1 ym and is of great value at 1.25 Fm [271. In the extreme it is possible through the use of thin membrane substrates, which effectively eliminate backscattered electrons, to produce linewidths as small as 250 A even using the standard electron resist PMMA. (See figure 5 and reference 1281.) Step and Repeat Reduction Electron Beam Projection In principle reduction electron beam projection systems 129, 301 are very much more efficient than scanning systems because millions of resolution elements are projected simultaneously. They are also simpler because pattern data are stored in a mask rather than fed on-line to the system during each exposure. Electron optical projection columns have been built which operate according to their design specifications, but fundamental problems remain with the mask and with distortion in the projected image. The difficulty with the mask is that there is no available substrate that is adequately transparent to electrons, so the mask must either be self-supporting or be a patterned photocathode. Thin metal foil masks have been fabricated and methods found to overcome the problem that certain areas of the pattern are unsupported; but adequate dimensional stability has not been demonstrated, nor have full complexity LSI masks been made. Photocathodes have relatively low electron brightness and would greatly increase exposure time over a thin foil mask illuminated with a high brightness electron beam. The electron optical system used to accelerate electrons from the photocathode surface has also been shown to increase pattern distortion [301. Distortion control has proven difficult over the large fields (l0,ooo times the mini- Fig. 5. Bright-field STEM image of pattern ionetched in 22.5 nm thick Au-Pd layer on 60 nm thick Si, N, membrane. Minimum lines and spaces are 25 nm wide. PMMA resist was exposed at 5 x Clem2 with a 50 KV I nm diameter electron beam. mum linewidth) needed to give these systems higher throughput than scanning systems. As with any step and repeat approach, the throughput of a reduction projection system will depend to a large extent on the time taken to mechanically step the sample and to register the pattern to the sample. With scanning systems writing is already becoming comparable to these overhead times. For projection systems to be faster, therefore, shorter exposure time is not adequate, the field size must also be larger so that the step and register sequence does not have to be performed so frequently. Field size for an optimum 4x projection system has been estimated to be 2 cm x 2 cm for 1 ym minimum linewidth (30). This is considerably higher than the maximum of about 0.5 cm x 0.5 cm predicted for scanning systems at the same resolution. Exposure time can be less than a second but distortion will be %. Scanning systems with dynamic distortion correction achieve % distortion. The relatively high distortion of the projection system may be tolerable for microcircuits provided all exposures are made on a single system, but it will preclude system to system compa- 708

6 tibility, or the possibility of combining this method with other lithography techniques. Electron beam proximity effects are d8icult to compensate with any full-pattern electron beam projection systems. Correction will have to be made by preadjusting the shapes in the mask because it will be difficult to adjust the electron charge per unit area from point to point in the image, as can be done with scanning electron beam systems. 1 : 1 Photocathode Electron Beam Projection This approach offers the resolution and depth of field of electron optics together with the parallelism and simplicity of proximity printing. Problems with the lifetime of the patterned photocathode delayed progress for many years but the introduction of cesium iodide as the photoemitter largely removed this problem. Successful methods for aligning the projected image to the sample have also been developed. The problems that remain are pattern distortion, reduction of image contrast because of backscattered electrons being returned to the sample surface, and correction for proximity effects. The sample is part of the electron optical system, and distortion is introduced if it is not absolutely flat. Calculations show that a 30 pm wafer bow will introduce a 1 pm error in pattern position [3 11. Electrostatic wafer chucking is being applied to solve this problem. So far 2 pm LSI devices have been successfully made with a 1 : 1 electron beam projector, and it is predicted that cm diameter wafers can be exposed in one hour [3 I]. For dimensions below 1 pm, the dimensional stability required to satisfy overlay requirements will become extremely severe just as it will for any full-wafer exposure technique. Demonstration of a method for correcting proximity effects by adjusting the shapes in the mask has been reported Electron Beam Proximity Printing Two methods [29, 321 have recently been reported in which flood electron beams are used for proximity printing through self-supporting 1 x masks. In the first case [29] the mask is illuminated with a parallel beam produced by the upper half of a reduction electron beam projection system column. A scanning mode in which the illuminating beam is focused onto the sample is proposed to test for alignment. The position of the shadow image will then be corrected before exposure by tilting the illuminating beam. 1.3 pm lines have been reproduced using a metal foil mask and an exposure time of 2.5 sec/cm2. In the second case 1321, relatively thick (3 pm) masks have been fabricated from silicon wafers by taking advantage of the high aspect ratios achievable with reactive ion etching of silicon. A gold layer absorbs the electrons. Exposure is made with a small electron beam (1 mm-2 mm in diameter) which scans over the mask. Linewidths of about 1 pm have been produced. For the fabrication of semiconductor devices a step and repeat approach will be used in order to keep the mask area relatively small and thereby make it easier to meet the required dimensional and overlay tolerances. A laser interferometer will keep track of sample position as the sample is stepped underneath the mask. Image position will be corrected at every chip site by tilting the illuminating beam. Mask fabrication for 1 x electron beam proximity printing will be more difficult than for electron beam reduction projection because of the smaller dimensions, but distortion should be easier to control, particularly in the second approach where a relatively small area mask is employed. The potential advantages of electron beam proximity printing over x-ray lithography are that exposure times can be shorter and that alignment can readily be achieved by electronically deflecting the illuminating beam. The self-supporting mask will be very much more difficult to fabricate, however, especially if shapes have to be adjusted to compensate for proximity effects. Electron scattering 709

7 Table 1. Alternative methods for employing new lithography approaches in the manufacture of microcircuits 5-lox MASTER MASK WORKING WAFER EXPOSE RETICLE MASKS ID PATTERN MECHANICAL CONTACT PATTERN REDUCTION GENERATOR CAMERA PROXIMITY PRINTING PROXIMITY PRINTING 2 PATTERN SCANNING DATA L E/B Ix OPTICAL PROJECTION lx E/B 4 L SCANNING E/B 5 PATTERN SCANNING X RAY DATA L El B - PROXIMITY - X RAY PROXIMITY from the edges of the mask openings may also degrade the image. Overall Comparison Table I shows five ways in which the various lithography methods can be combined to produce microcircuits. The first is the historically established route, the others are the methods most likely to appear in the future. Their relative success will obviously depend on requirements for device performance, cost and volume. A key will be the ratio of lithography cost to overall wafer cost. If overall wafer cost is high, then changing from a low cost lithography to a higher resolution higher cost approach may well reduce overall cost per device by increasing density. It is likely that combinations of the different approaches, for example, 2 and 4, may also prove optimum under some circumstances. Approximate estimates of comparative cost per resolution element for some of the different exposure systems are given in table 11. A resolution element is defined as a pattern square one minimum linewidth on the side. The numbers in brackets in the system column are predictions of field size, minimum linewidth and exposure time per field 710

8 ~ Table 2. Approximate estimates of comparative cost per resolution element for several lithography systems. SYSTEM PATTERN COST COST FACTOR PER PATTERN ELEMENTS ELEMENT PER SECOND (NORMALIZED TO F.F. OPTICAL PROJECTION) FULL-FIELD OPTICAL PROJECTION (70 cm, 2 pm, 25 Sec) 7 x 107 I 1 STEPIREPEAT OPTICAL PROJECTION 4.4 x (1 cm, 1.5 pm, 1 Sec) SCANNING ELECTRON BEAM 1.6 x (0,16 cm2, 1 pm, 1 Sec) X-RAY FULL FIELD (70 cm, 1 pm, 200 Sec) 3.5 x X-RAY STEPIREPEAT (1 cm2, 1 pm, 10 Sec) : 1 ELECTRON BEAM PROJECTION (70 cm, 1 pm, 60 Sec) 7 x IGNORING: YIELD TURN-AROUND TIME, MASK COST. OVERLAY ACCURACY. SY- STEM AVAILABILITY DATE, ETC (including overheads such as sample loading, mechanical stepping, and registration). That may be achievable in the future. In all cases a large degree of uncertainty is involved. Electron beam and x-ray systems can readily reach dimensions below 1 pm, but, in general, field size will be reduced and resolution elements per field will remain approximately equal. Electron beam reduction projection and proximity printing are not included because little if any device fabrication has been completed with these methods. In order to obtain the cost factor, the following capital costs were assumed: 1 : 1 scan- ning optical projection - $ 200 k; S/R reduction optical projection - $ 400 k, scanning electron beam $ 1500 k, x-ray full field $ 250 k, x-ray step and repeat $ 500 k (this could be the cost of one port on an electron synchrotron storage ring) $ 1250 k, 1 : 1 electron beam projection. Capital cost was distributed over five years and operating costs adjusted over a small range according to system complexity. One operator at a cost of $ 50 k per year was assumed for each system. According to table 11, a cost penalty will be incurred for all but one of the advanced approaches. This cost penalty will have to be 71 I

9 accounted for by increased density, higher yield, improved device performance, or on the basis of a logistical or cost advantage arising from the elimination of masks. References [ 11 Broers, A. N., Instit. of Phys. Conf. Series 40 (Instit. of Physics London), Ed. E. Ash, p. 155 (1978) and Broers, A. N., and T. H. P. Chang, Proc. Microcircuit Engineering 1978 Conf. (Cambridge University Press), Ed. W. C. Nixon and H. Ahmed. To be published 1979 [ 21 Lin, B. J., J. Vac. Sci. & Technol., 12, 1317 (1975) 3) Smith, H. I., Rev. Sci. Instr., 40, 729 (1969) 41 Spears, D. L., and H. I. Smith, Electron Lett., 8, 102 (1972) [ 51 Spiller, E., and R. Feder, X-ray Lithography, Chap. 3 in X-ray Optics, Ed. H. J. Queisser, Springer, Berlin (1977) 1 61 Spiller, E., et. al., J. Appl. Phys., 47, 5450 (1976) 71 Peckerar, M. C, et. al., Proc. 8th Internat. Conf. on Electron and Ion Beam Sci. Technol. (Electrochemical SOC. Princeton, N. J., USA.) To be published 1978 [ 81 Hatzakis, M., J. Vac. Sci. and Technol., 12, 1276 (1975) [ 91 Flanders, D., et. al., Appl. Phys. Lett., 31,426 (1977) [lo] Markle, D. A., Solid State Technol., p. 50 (June 1974) [111 Offner, A., SPSE 31st Annual Conf., J. Appl. Phot. Eng. (May 1978). Proc. to be published [ 121 Wilczynski, J. S., and R. Tibbetts, Unpublished (1976). See also Wilczynski, J. S., and R. Tibbetts, IBM J. Res. Develop., 13, 192 (1969) [131 Roussel, J., Sol. State Technol., p. 67 (May 1978) [ 141 Tobey, W., Electronics, 50, 115 (Aug ) [15] Hugues, M., et. al., Proc. Internat. Conf. on Microlithography, Paris, p. 71 (1977) [16] Yu, H. N., et. al., J. Vac. Sci. Technol., 12, 1297 (1975) [17] Reisman, A., et. al., and W. D. Grobmann, et a1 (1978). This Proceedings [ 181 Varnell, G., et. al., Proc. 6th Internat. Conf. on Electron and Ion Beam Sci. and Technol., Ed., R. Bakish, Electrochemical SOC., Princeton, N. J., p. 97 (1974) [191 Beasley, J. P., and D. G. Squire, IEEE Trans. on Electron Devices, ED-22, 3 (1975) [20] Herriott, D. R., et. al., IEEE Trans. on Electron Devices ED-22, 399 (1 975) [211 Ting, C. H., et. al., J. Vac. Sci. Technol., 15, 948 (1978) [22] Yourke, H. S., and E. V. Weber, Proc. Internat. Electron Devices Conf., Washington, D. C., p. 431 (1976) 1231 Munro, E., Optik, 39, 450 (1974) 1241 PfeiJfer, H. C., J. Vac. Sci., Technol., 15, 887 (1978) 251 Goto, E., et. al., J. Vac. Sci. Technol., 15, 883 (1978) 261 Wilson, A. D., et. al., Proc. 8th Internat. Conf. on Electron and Ion Beam Sci. and Technol., (Electrochemical SOC. Princeton, N. J., USA). To be published (1978) 1271 Parikh, M., J. Vac. Sci. Technol. 15, 931 (1978) 1281 Broers, A. N., et. al., Appl. Phys. Lett. (Sept. 1978) [291 Heritage, M. B., J. Vac. Sci. Technol., 12, 1135 (1975) [301 Lischke, B., et. al., Proc. Internat. Conf. on Microlithography, Paris, p. 163 (1977) I311 Scott, J. P., J. Vac. Sci. Technol., 15, 1016 (1978) [321 Bohlen, H., et. al., Proc. 8th Internat. Conference on Electron and Ion Beam Sci. Technol., (Electrochemical Society, Princeton, N. J., USA.). To be published 1978